Navigating the Shadows: A Comparative Analysis of SAR and Optical Imagery for Detecting (Dark) Vessels
(2025) In Master Thesis in Geographical Information Science GISM01 20251Dept of Physical Geography and Ecosystem Science
- Abstract
- Introduction
Illicit maritime activities frequently involve "dark" vessels that disable or manipulate their Automatic Identification System (AIS) signals to evade detection. This study evaluates the effectiveness of Synthetic Aperture Radar (SAR) and Optical imagery for vessel detection using both “manual” and automated Deep Learning (DL) methods while integrating AIS data to validate, improve detection accuracy and gain insights into vessel activities.
Methods
Using the Gulf of Lakonia, Greece, as a case study the research examines the capabilities of a SAR based Constant False Alarm Rate (CFAR) method, a SAR pre-trained detection model and an Optical imagery based Mask R-CNN Deep Learning (DL) approach for detecting vessels. The... (More) - Introduction
Illicit maritime activities frequently involve "dark" vessels that disable or manipulate their Automatic Identification System (AIS) signals to evade detection. This study evaluates the effectiveness of Synthetic Aperture Radar (SAR) and Optical imagery for vessel detection using both “manual” and automated Deep Learning (DL) methods while integrating AIS data to validate, improve detection accuracy and gain insights into vessel activities.
Methods
Using the Gulf of Lakonia, Greece, as a case study the research examines the capabilities of a SAR based Constant False Alarm Rate (CFAR) method, a SAR pre-trained detection model and an Optical imagery based Mask R-CNN Deep Learning (DL) approach for detecting vessels. The results are combined with AIS data in order to verify the satellite detections and potentially gain insights as to the vessels activities.
Key Results
The SAR CFAR method, relying on manual pre-processing and a calibrated CFAR detection algorithm, demonstrated a near perfect accuracy (F1 score: 100%) in detecting vessels, proving to be unaffected by factors like cloud cover. Conversely, the SAR pre-trained model exhibited a projection offset and lower detection accuracy (F1 Score: 67%) due to a projection offset, limiting its applicability and AIS integration. The Optical imagery based DL model achieved an F1 score of 90% with limitations arising from training dataset diversity, land/sea mask quality and cloud cover. Despite these challenges, Optical imagery provided valuable descriptive insights including vessel structure and color, which facilitated vessel identification (correlate imagery detection with a vessel’s name via the AIS). The integration of SAR and Optical imagery with AIS data enabled the detection of AIS manipulation and uncovered dark vessels engaged in Ship-To-Ship (STS) transfers. However, erroneous AIS data highlighted the necessity of multi-source approaches for result validation.
This study underscores the complementary nature of SAR and Optical imagery in maritime surveillance and highlights key challenges including the need for scalable automated solutions to improve detection accuracy and reduce reliance on manual processing.
Methodological Insights and Future Directions
The research demonstrates that integrating SAR and Optical imagery with AIS data can validate the detection results as well as provide vessel identification capabilities. While SAR imagery excels in detecting vessels, Optical imagery offers visual detail that aid vessel identification and classification. The study's findings look for further refinement of deep learning models, improved training datasets and enhanced integration techniques to minimize dependency on manual processes and account for erroneous data. Future research should focus on automating validation methods and develop Artificial Intelligence frameworks in scalable solutions. These have the possibility to assist authorities in strengthening their efforts to detect and deter illicit maritime activities. (Less) - Popular Abstract
- Illicit maritime activities frequently involve "dark" vessels that disable or manipulate their Automatic Identification System (AIS) signals (a vessel’s location, name, heading etc.) to evade detection. This study evaluates the effectiveness of using spaceborne images from two types of sensors, one active system (Synthetic Aperture Radar (SAR)) and one passive (Optical) for vessel detection. The methods refer to using both manual and automated (Deep Learning (DL)) approaches as well as integrating them with AIS data in order to improve detection accuracy and gain insights into vessel activities. Using the Gulf of Lakonia, Greece, as a case study, it examines the capabilities of a SAR based “manual” approach and a pre-trained DL detection... (More)
- Illicit maritime activities frequently involve "dark" vessels that disable or manipulate their Automatic Identification System (AIS) signals (a vessel’s location, name, heading etc.) to evade detection. This study evaluates the effectiveness of using spaceborne images from two types of sensors, one active system (Synthetic Aperture Radar (SAR)) and one passive (Optical) for vessel detection. The methods refer to using both manual and automated (Deep Learning (DL)) approaches as well as integrating them with AIS data in order to improve detection accuracy and gain insights into vessel activities. Using the Gulf of Lakonia, Greece, as a case study, it examines the capabilities of a SAR based “manual” approach and a pre-trained DL detection model. At the same time a similar approach is used with an Optical imagery based DL detection model. The results are combined with AIS data in order to verify the satellite detections and potentially gain insights as to the vessels activities.
The SAR “manual” processing method, demonstrated near perfect accuracy in detecting vessels, proving to be resilient against various parameters (e.g. cloud cover). Conversely, the SAR pre-trained model exhibited a projection offset and lower detection accuracy limiting its applicability and was finally excluded from the AIS integration. The Optical imagery DL model presented just 10% lower accuracy in comparison to SAR, with limitations arising from the training dataset’s diversity, land/sea mask quality etc. Despite these challenges, Optical imagery provided valuable descriptive insights like structure and color which better facilitated vessel identification. The integration of the detections with AIS data uncovered AIS manipulation and potential “dark” vessels. However, erroneous AIS data highlighted the necessity of multi-source approaches for result validation.
The study underscores the complementary nature of SAR and Optical imagery in maritime surveillance. It highlights the key challenges and the need for automated solutions to improve detection accuracy, identify erroneous data and reduce reliance on manual processing.
The study's findings show the need for improved training datasets and multi-sourced approaches. Future research should focus on automating validation methods using Artificial Intelligence driven frameworks. By implementing scalable solutions, authorities can strengthen efforts to detect and deter illicit maritime activities more effectively. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9186636
- author
- Mourampetzis, Athanasios Emmanouil
- supervisor
-
- Lina Eklund LU
- organization
- alternative title
- A Comparative Analysis of SAR and Optical Imagery for Detecting (Dark) Vessels
- course
- GISM01 20251
- year
- 2025
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Geography, GIS, Synthetic Aperture Radar (SAR), Constant False Alarm Rate (CFAR), Optical Imagery, Automatic Identification System (AIS), Mask Regional Convolutional Network (Mask R-CNN), Maritime Surveillance, Dark Vessels
- publication/series
- Master Thesis in Geographical Information Science
- report number
- 187
- language
- English
- id
- 9186636
- date added to LUP
- 2025-03-18 16:26:33
- date last changed
- 2025-03-18 16:26:33
@misc{9186636,
abstract = {{Introduction
Illicit maritime activities frequently involve "dark" vessels that disable or manipulate their Automatic Identification System (AIS) signals to evade detection. This study evaluates the effectiveness of Synthetic Aperture Radar (SAR) and Optical imagery for vessel detection using both “manual” and automated Deep Learning (DL) methods while integrating AIS data to validate, improve detection accuracy and gain insights into vessel activities.
Methods
Using the Gulf of Lakonia, Greece, as a case study the research examines the capabilities of a SAR based Constant False Alarm Rate (CFAR) method, a SAR pre-trained detection model and an Optical imagery based Mask R-CNN Deep Learning (DL) approach for detecting vessels. The results are combined with AIS data in order to verify the satellite detections and potentially gain insights as to the vessels activities.
Key Results
The SAR CFAR method, relying on manual pre-processing and a calibrated CFAR detection algorithm, demonstrated a near perfect accuracy (F1 score: 100%) in detecting vessels, proving to be unaffected by factors like cloud cover. Conversely, the SAR pre-trained model exhibited a projection offset and lower detection accuracy (F1 Score: 67%) due to a projection offset, limiting its applicability and AIS integration. The Optical imagery based DL model achieved an F1 score of 90% with limitations arising from training dataset diversity, land/sea mask quality and cloud cover. Despite these challenges, Optical imagery provided valuable descriptive insights including vessel structure and color, which facilitated vessel identification (correlate imagery detection with a vessel’s name via the AIS). The integration of SAR and Optical imagery with AIS data enabled the detection of AIS manipulation and uncovered dark vessels engaged in Ship-To-Ship (STS) transfers. However, erroneous AIS data highlighted the necessity of multi-source approaches for result validation.
This study underscores the complementary nature of SAR and Optical imagery in maritime surveillance and highlights key challenges including the need for scalable automated solutions to improve detection accuracy and reduce reliance on manual processing.
Methodological Insights and Future Directions
The research demonstrates that integrating SAR and Optical imagery with AIS data can validate the detection results as well as provide vessel identification capabilities. While SAR imagery excels in detecting vessels, Optical imagery offers visual detail that aid vessel identification and classification. The study's findings look for further refinement of deep learning models, improved training datasets and enhanced integration techniques to minimize dependency on manual processes and account for erroneous data. Future research should focus on automating validation methods and develop Artificial Intelligence frameworks in scalable solutions. These have the possibility to assist authorities in strengthening their efforts to detect and deter illicit maritime activities.}},
author = {{Mourampetzis, Athanasios Emmanouil}},
language = {{eng}},
note = {{Student Paper}},
series = {{Master Thesis in Geographical Information Science}},
title = {{Navigating the Shadows: A Comparative Analysis of SAR and Optical Imagery for Detecting (Dark) Vessels}},
year = {{2025}},
}